Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- AWS announced GA October 5, 2026 for this speech-to-speech model targeting voice support, assistants and real-time conversation.
- The listed regions are N. Virginia, Oregon, Stockholm and Tokyo, not all AWS regions.
- Improved reasoning, instructions, tool accuracy and latency are AWS claims, not our measurements or a uniform speedup percentage.
- The announcement says the same price as Nova 2 Sonic, but the fetched pricing page returned dynamic placeholders; no rate or free allowance is invented here.
- The announcement says 256K context and current model page up to 1M. Resolve version-specific limits before promising capacity.
- Keep configuration and evaluation baselines, and validate voice/text transitions and tool results. Upgrades replace neither business permission, human confirmation nor quotas.
What this release changes
AWS's October 5 announcement, Announcing Amazon Nova 2.5 Sonic with improved reasoning for voice agents, establishes GA rather than a future preview. It takes speech and returns speech for continuous conversation, not an upgrade of every Nova text model.
AWS emphasizes multi-step instructions, context and connected tools, using order lookup, return-eligibility checks and initiating a return as an example. That demonstrates intended use, not a connection to your order system or authorization to bypass business rules.
Check the four regions first
The announcement lists Bedrock in US East N. Virginia, US West Oregon, Europe Stockholm and Asia Pacific Tokyo. It does not announce every region. Compare your current application region, then inspect its actual model entry and invocation configuration.
Regional availability, account authorization and client compatibility differ. Public release proves no account permission, capacity or existing project configuration. If a version is missing, retain region, model name and documentation version for investigation rather than copying an old ID or blindly changing business regions.
Pricing parity is not identical conversation cost
AWS says Nova 2.5 Sonic pricing matches Nova 2 Sonic. At the October 6 approximately 08:19 Shanghai check, speech/text input/output rates were dynamic placeholders and the table used the older name. No directly verifiable 2.5 numeric rates were obtained, so only the official parity statement is retained.
Parity does not guarantee the same total per conversation. Record actual model, region, input/output modalities and usage, then check applicable pricing. Compare equivalent tasks, conversation lengths and tools rather than converting lower-latency claims into guaranteed bill reductions.
Do not mix context and version documentation
The announcement mentions inherited 256K context, while the current Nova page says up to 1M for 2.5 Sonic. No specific limit explanation reconciled them at sampling. Do not promise 1M capacity or imply full upgrade conditions were published.
Both pages mention seven languages. The fetched specialist guide still says Nova 2 Sonic and lists English, French, Italian, German, Spanish, Portuguese and Hindi, not a complete updated 2.5 table. Chinese speech, voices, model ID, parameters and connection limits require corresponding 2.5 evidence.
The older Nova Sonic page explicitly labels itself Version 1 and links to Nova 2 Sonic. Check version banners before adopting old language counts, events or limits as new-model rules.
Begin without real business writes
Keep the working application region, model configuration and version. List voice continuity, interruption recovery, voice/text switching and tool-result return as acceptance checks. This is a preparation sequence; no model call or account verification occurred here.
Use fictional nonsensitive questions and read-only or simulated tools first. For example, return a fictional order status and check faithful explanation rather than connecting real returns, payments, email or customer data. Fluent speech alone proves neither tool execution nor satisfied business conditions.
Judge complete results, not one audio response
Define success before replacing the current version. Check instruction completion, errors/timeouts, tool arguments/results and interrupted-session behavior as well as audible output. Real operations retain existing permissions and confirmation steps; improved tool accuracy replaces none of them.
If only some steps succeed, retain working configuration and comparison evidence, stop expansion and check model version/docs. Do not loosen permissions, expose keys or repeat uncertain writes to obtain one success. Record account capacity, region and networking errors separately from the release.
Track confirmed release separately from missing details
Confirmed facts are GA, four regions, voice-agent focus and stated parity pricing. The announcement also says Strands Bidi Agents is GA, which does not mean your app installed or configured it.
Sources were read October 6 between 00:16 and 00:20 UTC, or 08:16–08:20 Shanghai time, not an exact first-publication second. Prioritize 2.5 ID, version-specific languages/connections, context reconciliation and verifiable regional prices. Missing details do not negate GA, but still affect configuration and capacity planning.
Sources
Frequently Asked Questions
Is Nova 2.5 Sonic preview or GA?
The October 5 AWS announcement says GA in four Bedrock regions. Check your account permissions and client configuration separately.
Which regions are listed?
N. Virginia, Oregon, Stockholm and Tokyo. Do not infer simultaneous availability elsewhere.
Is it more expensive than Nova 2 Sonic?
AWS says pricing is the same. The fetched page yielded no verifiable 2.5 numeric rates. Actual total depends on model, region, modalities and usage.
Is context 256K or 1M?
Official pages differ: 256K in the announcement, up to 1M on the model page. No explanation was verified; use version-specific limits before capacity commitments.
Can I use Chinese speech or an old model ID?
Not on this evidence alone. Specialist docs still identify Nova 2 Sonic; a complete 2.5 language table, ID and parameters were not established.
Will my voice-support app upgrade automatically?
The announcement does not say so. Keep current configuration and verify target region/version using nonsensitive, non-writing trials.